مجله پردازش سیگنال پیشرفته
سال پنجم شماره 1 (پیاپی 7، بهار و تابستان 1400)

It is possible to noticeably increase the capacity of wireless communication systems through the use of multiple antennas both in the transmitter and in the receiver. In such systems, which are referred to in short as MIMO, the receiver uses its knowledge of the channel to detect the transmitted signal. Different methods have been proposed for optimal and sub-optimal detection of the transmitted signals. Recently, principles of deep learning and implementing neural networks have been employed as a near optimal approach for MIMO detection with fewer calculations during the testing process compared to traditional methods. In the event an error occurs in the receiver’s channel estimation process, this type of detector suffers a drop in performance and as a result, BER will increase. Given that in practice, the receiver only has an estimation of the CSI instead of the exact values, the current study presents an enhanced detection method based on deep learning, which is also robust against channel estimation error. In this detection method, by using the covariance matrix of the channel estimator and the principles of deep learning, a robust detector against channel estimation error is proposed and comprehensively evaluated. Numerical simulations confirm the performance of the proposed method.

Among the techniques that are used for signal denoising, smoothing filters have received significant attention during the past. However, these methods are particularly suited for polynomial signal smoothing. Therefore, their performance is significantly decreased for signals that cannot be well modelled with a polynomial function. To overcome this limitation, in this paper, we propose a new approach to smoothing filter design, which is based on the delay differential equation model. In this approach, we propose to substitute the derivative of the signal with a DDE model of the signal. As an example, a delay differential equation of moving average (MA) model is used as penalty term in the optimization problem. The results indicate that a better solution can be found by appropriate balancing a trade-off between the MA model of the signal and the minimum mean square error. The proposed MA smoothing filter is analyzed in frequency domain. It is shown that the proposed MA smoothing filter displays good properties within its pass-band and stop-band bands for small values of window length. As an application, the proposed MA smoothing filter was used for electrocardiogram (ECG) signal denoising. We tested the method over data from the PhysioNet PTB database. The results show that the proposed MA smoothing filter outperforms the original smoothness priors or QV regularization.

Detection of calcium particles in breast mammography images is important in the early detection of cancer. Identification of these particles is done manually by experts, which is associated with high cost and error. In this paper, a new method based on fuzzy clustering algorithm for fine-grained detection in mammographic images is proposed. In the proposed method, the low quality of mammographic images is improved with the help of preprocessing. By defining an appropriate membership function in fuzzy clustering, fine-grained regions are identified. Finally, the identified areas were classified into benign and malignant groups with the help of forward propagation neural network with error propagation training algorithm. The accuracy of identification of the desired area is 96.79% and the sensitivity of this identification is 97.20%. Compared to the previous method, the accuracy and sensitivity of fine-grained identification has been improved (95% of the desired area identification accuracy and 90.52% sensitivity).  In the classification of areas with the help of neural network, the accuracy was 97.5%. Evaluation criteria showed the superiority of the proposed method in the extraction of calcium particles and classification. The reason for the superiority of the proposed method is the high accuracy in extracting the desired area as well as the distinctive features extracted from the desired area.

Pulse-Doppler radars typically use pulse compression and Doppler processing to detect moving targets through fast Fourier transforms. The conventional pulse compression method and the standard matched filter output for detection of small targets close to a large target do not work well, since the sidelobes of the match filter output by a large target could mask the smaller targets. Adaptive pulse compression resolves this issue significantly in noise. However, the fast targets induce Doppler phase shift in the received signal frequency, in which cause mismatch between the received signal and the transmitted signal. Consequently, the Signal to Noise Ratio is reduced. Whereas the matched filter in the radar receiver is only adapted to the transmitted signal version and its output will be wasted due to non-matching with the received signal from the environment. Adaptive pulse compression is generally applied with a single pulse in alone noise environment, but in the presence of strong clutter it is required to several return pulses. In this paper, to supply these pulses, in a radar transmitter equipped with adaptive pulse compression, waveforms diversity are generated by random frequency hopping in step frequency waveform. The simulation results of the detection of masked moving targets are compared with other conventional methods.

Breast cancer has increased among women in recent years and is one of the leading causes of death in women. Studies show that thermography is a faster, cheaper, passive, risk-free, radiation-free and pain-free method than other diagnostic methods. New methods of image processing, vision and machine learning have led to successful investigations into the invention of breast cancer detection systems by thermometric images. In the present study, a proper method of diagnosing abnormality through thermography images of the obverse view is presented. By this segregation method, the breast area and every other area targeted by the physician that is vital for breast cancer diagnosis are color-divided in the thermographs. Warmer regions known as vital centers are extracted by the FCM algorithm and the fractal dimension of these regions is calculated using three different methods. The Studies suggesting that fractal analysis may potentially improve the reliability of thermography in breast tumor detection. The innovative aspect of this paper is the study of the role of fractal analysis in tracking the symmetrical heat distribution in two breast tissues in thermographic images. The results show that fractal analysis plays an important role in tracking the symmetrical heat distribution in two breast tissues to investigate asymmetry in order to detect breast abnormalities.

In current cellular systems, the performance of active users' devices at the cell edge suffers from the poor link quality. However, these connections also requires more resource blocks and transmission power. In order to reduce the number of resource blocks and transmission power, this paper discusses device to device communication in downlink and uplink cases of cellular communication systems. In order to optimize the connections of different network users, which means finding the best user’s connection to a base station (minimum power consumption), which may be established through communication with other users or direct connection with the base station, and to minimize the total transmission power, different optimization methods such as gravitational search optimization, particle swarm optimization, genetic optimization algorithm and distributed strategy based on Q learning and softmax decision making methods are used. The numerical results show a power reduction of around 30 percent for these distributed communications with less computational complexity using the Q learning method compared to the case in which all users traditionally connect through the base station in a centralized way with high computational complexity.

In this paper, a CODEC system based on super resolution, is proposed for compression of high resolution textual images. It employs image resizing to decrease image dimensions and consequently, to improve the compression ratio; but at the other hand, it may reduce the image quality. Therefore, the decompression unit employs super resolution to simultaneously increase the reconstructed image dimensions and quality. In the employed interpolation-based super-resolution method, using an efficient textual image matting algorithm, the input low-resolution textual image is decomposed into three layers after which, each layer is enlarged using a particular method. Finally, the enlarged layers are combined to build the high resolution reconstructed textual image. An interesting feature of the proposed method is the ability to use existing compression methods such as JPEG, JPEG2000 and SPIHT. We have employed the aforementioned compression methods in the proposed CODEC system and evaluated the compression results with respect to OCR rate, Mean Opinion Score (MOS), and PSNR measures. Considering the OCR and MOS measures, the proposed method outperformed the others but not so with respect to PSNR.

One of the non-orthogonal access techniques is non-orthogonal multiple access (NOMA) that is known as an access technique with high spectral efficiency. The combination of NOMA technique and MIMO system which is called MIMO-NOMA system has been offered in order to achieve higher capacity and spectral gain for 5G mobile communication. The performance of communication systems especially MIMO-NOMA system, significantly depends on channel estimation accuracy as the system efficiency decrease dramatically in the presence of channel estimation error. In this paper, the channel estimation error effect in downlink MIMO-NOMA system is investigated and a new detector which aims to improve the performance of system is proposed despite the channel estimation error. The proposed detector decreases both the effects of channel estimation error and the interference of user in three steps. Simulation results indicate that the proposed detector decreases the error probability and improves the system performance compared to that of the MMSE detector for far user and MMSE-SIC detector for near user considerably.

This paper deals with locating disruptive sources in patients with brain disorders, supposing to have the location of active brain sources in healthy people according to their functional connectivity pattern information in similar activities. In the proposed algorithm, firstly the effect of sources that are active in normal brain activity is eliminated from the patient’s recorded EEG signals using the LCMV beamformer. Then, the disruptive sources are localized. The proposed method utilizes a combination of Sequential Particle Filter (SPF) and LCMV Beam-Former (BF) to localize disruptive sources. The simulations have been performed using BrainStorm software and pseudo-real EEG signals. The results of applying the proposed method (SPF-BF) on the simulated EEG signal show that this method could achieve better results in severe noise conditions than the LCMV beamformer, traditional particle filter algorithms, and combination of them. Also, the comparative results of the proposed method and sLORETA confirm the proper performance of the proposed method. In addition, the proposed method outperforms the other methods in terms of computational complexity.

Researchers are trying to achieve the power of the human mind by implementing decision-making algorithms similar to brain function. Hierarchical decisions are complex decisions that require metacognitive reasoning mechanisms in the brain. Negative feedback, certainty, and motion strength are the parameters that play a role in shaping such decisions. In this study, in order to design a computational framework similar to brain function for intelligent systems, it will be important to understand the biology nature of high-level decision-making, using other types of data in addition to behavioral data. Since involuntary eye responses resulting from the output of psychophysical experiments are a reliable representative of the function of the neuronal mechanism in the brain, in this study addition to the analysis of behavioral data, this issue has been addressed whether it is possible to understand the dynamics of changes in high-level decisions by analyzing involuntary human data (eye signals). We found that pupil diameter size predicts the likelihood of changes in the parameters of high-level decisions, and reflects the individual's high-level decision strategy under complex conditions. Then, in order to design systems similar to brain function in complex environments, we provide a framework for hierarchical decisions.

Due to the epidemic of the coronavirus (Covid-19) and its rapid spread around the world, the world has faced a huge crisis. To prevent the spread of the coronavirus, the World Health Organization (WHO) has introduced the use of masks and keeping social distance as the best preventive method. So, developing an automatic monitoring system for detection of facemask in some crowded places is essential. To do this, we propose a mask recognition system based on transfer learning and Inception v3 architecture. In the proposed method, two datasets are used simultaneously for training including: Simulated Mask Face Dataset (SMFD) and MaskedFace-Net (MFN).this paper tries to increase the accuracy of the proposed system by optimally setting hyper-parameters and accurately designing the fully connected layers. The main advantage of the proposed method is that in addition to masked and unmasked face, it can also detect cases of incorrect use of mask. Therefore, the proposed method classifies the input face images into three categories. Experimental results show the high accuracy and efficiency of the proposed method; so that, this method has achieved to accuracy of 99.47% and 99.33% in training and test data respectively.